Microsoft is working on its own AI processor for training large models. This is what sources tell tech site The Information.
According to the website, the tech giant has been developing its own AI processors since 2019 under the name Athena. Meanwhile, copies of the processors are reportedly already available to a limited number of Microsoft employees and AI developer OpenAI. The employees are currently testing the processors.
The hardware chips will be used specifically for training Large Language Model (LLM) AI models, such as GPT-4. They should also better help combat interference around the deployment of AI models.
Alternative to other vendors
Microsoft’s development of its own AI processors is said to be prompted by the tech giant’s search for alternatives to the AI-enabled processors it currently buys from other vendors. These include Nvidia’s A100 and H100 accelerators, which are widely used for AI model training.
With its own AI processors, Microsoft expects to save money and time for its AI developments, writes The Information. Microsoft is not the only major hyperscaler developing its own processors to be less dependent on other vendors. AWS, Google and Facebook are also making their own processors.
Whether the proprietary AI processors will at some point be offered to a wider audience, such as for Azure users, is not known. Perhaps they are intended only for internal use. Microsoft does reportedly have its own internal roadmap for these specific processors.
Also read: OpenAI introduces GPT-4, also available in Microsoft Bing